ExpertsFMTSMC (Remastered)
Acquired · January 21, 2025
5 Selected Lessons:
framework · 2 case · 1 reasoning_pattern · 1 heuristic · 1
Learning-curve pricing to capture market share and margin
Source
Acquired — TSMC (Remastered) (January 21, 2025) · at 30:28 ▸
When this applies
when entering a capital-intensive production business where yields/costs improve with cumulative volume
The takeaway
In businesses with a learning curve and high fixed costs, price to maximize volume early rather than to recoup costs immediately, because volume itself is what reduces costs and builds market share.
Why it works
Pricing low from the start drives volume, which accelerates the yield-improvement learning curve and keeps the plant running at max capacity, converting fixed costs into highly profitable marginal output faster than high initial pricing would.
One level up
Counterintuitive near-term sacrifice (margin) can compound into a durable long-term advantage (share plus lower costs) when the underlying cost structure rewards scale.
Where it stops holding
This stops working when there's no learning curve (costs don't fall with volume) or when competitors can match low pricing without needing to also build the same fixed-cost base.
Detail
procedure:
- Recognize that new process technology starts at low yield/high cost but improves with volume
- Price low from product launch rather than pricing to recover fixed costs immediately
- Continually and automatically reduce price every quarter even without market pressure to do so
- Use the resulting demand surge to reach max capacity utilization fast, accelerating the yield learning curve
- Capture market share from competitors who are pricing high
benchmarks: {"result":"TI's IC business became the biggest and most profitable in the world"}
rationale: Fixed costs are already spent; the benefit is in running at capacity and owning the market before competitors catch up.
In their words
The result was so-called learning curve pricing. So start low and then continually automatically reduce the price every quarter, even when the market did not demand it. And this was a very successful effort, even though it was somewhat controversial.
Seven Powers applied to TSMC
Source
Acquired — TSMC (Remastered) (January 21, 2025) · at 1:53:20 ▸ · at 1:55:35 ▸
When this applies
when evaluating any company's durable competitive advantage by systematically checking it against a known taxonomy of power sources
The takeaway
To evaluate whether a company's moat is real and durable, systematically test it against each known category of competitive power rather than relying on a single favored explanation.
Why it works
Explicitly walking through each of the seven named sources of defensibility forces a structured, non-cherry-picked assessment, revealing which powers actually apply (here: counter-positioning, scale economies, switching costs, and especially process power) versus which don't (branding, network economies, cornered resources).
One level up
Durable advantage is best understood as a portfolio of distinct mechanisms, not a single 'moat' — and multiple reinforcing powers compound into near-impenetrability.
Where it stops holding
This stops working when a company's situation includes power sources (like direct geopolitical/regulatory risk) that don't map cleanly onto the existing seven categories, as the hosts found when trying to classify China/Taiwan risk.
Detail
procedure:
- Counter-positioning: check if the incumbents' own business model prevents them from copying the challenger's model (IDMs' brand-and-market demands prevented a pure foundry model)
- Scale economies: check if larger scale directly reduces cost per unit
- Switching costs: check if customers are so deeply integrated into your process that leaving is costly (TSMC customers can't easily port to Global Foundries)
- Network economies: check if more customers directly benefit other customers (ruled out for TSMC, unlike social platforms)
- Process power: check if decades of accumulated know-how can't be replicated even with equal capital (judged TSMC's strongest power)
- Branding and cornered resources: check applicability (ruled out/noted as largely irrelevant for TSMC)
benchmarks: {"conclusion":"Process power and scale economies judged TSMC's top two sources of defensibility"}
rationale: Systematic framework application surfaces non-obvious conclusions, like network economies not applying despite TSMC's deep ecosystem integration.
In their words
To me, this is the clearest example I could ever imagine of process power. It takes all 40 years of TSMC's history to have arrived at where they are today. And even if 10 people left and tried to start the next TSMC
Becoming the platform instead of betting on the players
Source
Acquired — TSMC (Remastered) (January 21, 2025) · at 1:14:13 ▸
When this applies
when deciding whether to bet on individual future winners in an emerging space versus building the infrastructure/platform that all eventual winners will depend on
The takeaway
When you can't predict which specific players will win in an emerging market, consider becoming the shared infrastructure that all of them need, rather than betting selectively on a handful.
Why it works
By becoming the enabling infrastructure rather than picking individual winners, Chang captured value from the entire ecosystem of future entrants regardless of which specific company succeeded, instead of concentrating risk/reward on a few bets.
One level up
Infrastructure-layer positioning diversifies and compounds exposure to an ecosystem's growth better than picking individual winners.
Where it stops holding
This stops working when the infrastructure layer requires more capital or capability than you can muster, or when the ecosystem never materializes at all.
Detail
approach: Instead of investing directly in individual startups you believe will succeed, build the platform/service layer that any successful entrant in that space will need.
signals:
- Emerging category with high capital barriers to entry for individual players
- You have insider knowledge the category wants to exist but structurally can't
- You can position yourself as the enabling layer rather than a competitor
rationale: TSMC effectively 'indexed' the entire future fabless semiconductor industry (NVIDIA, Qualcomm, Broadcom, Marvell) by being what they all needed, rather than picking a subset to invest in.
In their words
He takes an index out on the whole future innovation and entrepreneurship market in semiconductors by becoming the platform that they're going to build on instead of like going and investing in them. Like he enables all of it. He's like the Y combinator of semiconductors.
Moore's Law and Rock's Law combine into winner-take-all capital dynamics
Source
Acquired — TSMC (Remastered) (January 21, 2025) · at 1:49:46 ▸
When this applies
when assessing competitive dynamics in any industry where both (a) operating costs fall with scale, and (b) fixed capital costs to enter/compete rise over time
The takeaway
When both unit economics (favoring scale) and entry costs (rising over time) move in the same reinforcing direction, expect the industry to consolidate into a true winner-take-all/few structure, not just an oligopoly.
Why it works
Moore's Law describes falling operating costs per unit at scale (benefit of production), while Rock's Law describes exponentially rising fixed capital costs to build the next-generation plant (barrier to entry); together they mean only the largest, most profitable incumbent can afford to keep building the next generation of capacity, locking out new entrants and even established competitors.
One level up
Two independent laws that reinforce rather than conflict with each other create structurally durable advantage, stronger than either alone.
Where it stops holding
This stops working when a paradigm shift changes the underlying substrate/technology such that the accumulated capital advantage no longer applies.
Detail
rule: Rising fixed capital costs for cutting-edge capacity (Rock's Law) combined with falling operating costs at scale (Moore's Law) produce winner-take-all dynamics.
threshold: TSMC's CapEx growing from $17B to $60B in two years — faster than Rock's Law's four-year doubling — signals accelerating consolidation.
rationale: Only the already-dominant, highly profitable player can keep funding ever-more-expensive fabs, excluding competitors structurally, not just financially.
In their words
Well, shoot, if we just look at TSMC's CapEx forecasting, they're going from 17 to 30 to 60 over two years. So that's way faster than four years.
Betting on a market that doesn't exist yet (build it and they will come)
Source
When this applies
when founding a platform business on the belief that enabled customers will emerge once the platform exists, even though no current market validates it
The takeaway
A 'build it and they will come' bet can be rational, not reckless, if you possess privileged, direct evidence of latent suppressed demand that outsiders can't see — but the hosts note you cannot know in advance, from the strategy alone, whether it will fall into the roughly one-in-ten cases that succeed or the nine that fail.
Why it works
Chang had first-hand observational evidence (designers at TI/GI wanting to leave but blocked by capital needs for fabs) that gave him a differentiated, insider reason to believe a market would emerge, even though outsiders saw no market and it was conventionally considered a bad strategy.
One level up
Outcome survivorship bias makes it impossible to retroactively judge whether a bold be-the-platform strategy was 'good' independent of whether it happened to work.
Where it stops holding
This stops working when the founder lacks genuine insider insight into suppressed latent demand, or when, as conventional investors (e.g. Don Valentine) argue, you should only invest where a market already exists rather than trying to develop one.
Detail
situation: TSMC was founded as a pure-play foundry with no fabless chip industry yet existing to serve.
decision: Morris Chang proceeded anyway, based on his insider observation that designers wanted to start companies but were blocked by fab capital requirements.
outcome: The fabless industry (Qualcomm, NVIDIA, Broadcom, etc.) did eventually emerge and TSMC became indispensable — but the hosts explicitly flag this could have failed.
lesson: If you're only looking at outcomes that happened, you cannot reverse-engineer the true probability that justified the bet — bold market-creation strategies are inherently unfalsifiable in hindsight.
In their words
My next one is that if you're only looking at the outcomes that happened, you cannot reverse engineer what the probability that Oda happened is. And this is a very abstract way of me saying the strategy of if you build it, they will come that Morris implemented is a bad strategy.